Applied Thermal Engineering 195 (2021) 117160

Contents lists available at ScienceDirect

Applied Thermal Engineering

journal homepage: www.elsevier.com/locate/apthermeng

Exploring on a three-fluid Eulerian-Eulerian-Eulerian approach for the

prediction of liquid jet atomization
Xiaohang Qu a, Shripad Revankar b, Xiaoni Qi a, *, Qianjian Guo a
a
Department of Energy and Power Engineering, Shandong University of Technology, Zibo 255000, PR China
b
School of Nuclear Engineering, Purdue University, West Lafayette, IN 47907, USA

A R T I C L E I N F O A B S T R A C T

Keywords: Predicting the atomization of a liquid jet in its applications, such as in fuel combustion and nuclear safety systems
Jet atomization and in many other critical industrial applications, remains a challenging task. Using a low-computer consump­
Three-fluid approach tion coarse-grid, this work presents a 3-D numerical method with three phases, which respectively represent the
Multi-scale
gas, continuous liquid and dispersed liquid. A Eulerian description is used for each of these phases. In addition,
AIAD-model
Population balance model
an algebraic interfacial area density (AIAD) model is used to consider the continuous liquid and gas phases.
CFD Meanwhile, a discrete population balance model is applied in order to take the droplet breakup and coalescences
into account. The present Eulerian approach is then tested for the different liquid jets used in atomization re­
gimes and then this is validated against the experimental data. The comparison reveals reasonable agreement in
aspects such as jet spreading, droplet coalescence and the distribution of droplet sizes. Based on the simulation,
the disintegration rates are found to increase from 175 kg/m3/s for case 1 to 310 kg/m3/s for case 2, which is due
to increases in ejection velocity and turbulence intensity. In both the simulation and the experiment, larger-sized
droplets form as the jet evolves downstream. This indicates that the coalescence between droplets overwhelms
the possible breakup, meaning therefore that the diameter of the droplets increases streamwise, as shown in the
comparison between D10 and D32 at the axial positions of 200 mm and 400 mm.

1. Introduction operation and the short characteristics of time and length. From an
experimental point of view, the measurable characteristics are exceed­
An important prerequisite for the development of technologies for ingly limited in the above-mentioned adverse conditions. Therefore, it is
thermal energy production is the creation of efficient engines that utilize important to develop a new method that aims to create a precious
liquid fuel to generate heat. The problem of achieving an effective simulation of the liquid–gas jet flow.
dispersion of liquid–gas fuels, meanwhile, is one of the key points to be In recent years, researchers have employed various methods to
considered when designing an engine [1]. The quality of fuel dispersion simulate the breakup of liquid jets. Most of these studies have been
affects the mixing efficiency and ignition stability, while also deter­ based on a single fluid interface tracking method, such as the volume of
mining the heat generation, completeness of fuel burnout, rate of com­ fluid (VOF) [2], level set [3] or a combination of both [4]. Though these
bustion, and the toxic composition of the combustion emissions. Liquid- interface tracking methods require less experimental closure relations,
gas jets are employed in many applications including liquid fuel com­ their precise depiction of the interfacial forces requires a highly resolved
bustion, chemical technology and medical sprays, firefighting, as well as interface and turbulent structures. Thus, highly accurate turbulent
being used for many other critical industrial applications. The study of models such as the direct numerical simulation (DNS) [5,6] and large
these liquid–gas jets has always been difficult since their applications are eddy simulation (LES) [7] are needed, and these are very computa­
usually characterized by complex phenomena including the atomization tionally expensive. Salvador et al. [8] performed a Gerris-based DNS
of the jet into smaller droplets, its mixing with atmosphere fluid, its simulation of a low injection pressure diesel spray, and the results
coalescence of disintegrated droplets, its transfer of mass and mo­ indicated good agreement with the theoretical model, in regards to the
mentum, and its evaporation. This complexity is accentuated in the case axial velocity aspect. Due to the limitations imposed by the computer’s
of the fuel jets used in engines due to the high frequency transient limited capabilities, only the primary atomization region, which covered

* Corresponding author.
E-mail address: [email protected] (X. Qi).

https://doi.org/10.1016/j.applthermaleng.2021.117160
Received 21 February 2021; Received in revised form 12 May 2021; Accepted 26 May 2021
Available online 31 May 2021
1359-4311/© 2021 Elsevier Ltd. All rights reserved.
X. Qu et al. Applied Thermal Engineering 195 (2021) 117160

a zone very close to the nozzle, was simulated, while the number of cells liquid droplets is superior in its efficiency to the Lagrangian tracking
reached 12.8 million. Using a combined mass conservative level set method. However, using interface tracking to resolve all the interface
method as well as a ghost fluid method, Shao et al. [9] were able to structures between the Eulerian continuous liquid and the gas phases
numerically show the evolution of a fuel jet’s swirling primary atomi­ could still be very expensive. Since in practice the flow characteristics of
zation. However, only a cubic zone of 4 mm was able to be calculated the two-phase flow matter as a whole, as opposed to simply as the
with a down-scaled Reynolds number and Webber number, while more specific locations and movements of interfaces, it is sometimes unnec­
than one billion cells are needed in order to resolve the Kolmogorov essary to resolve the detailed structures between the continuous liquid
scale of turbulence. With the help of a Tianhe-1A super computer and gas phases. This is particularly true when the two-phase flow occurs
located in Changsha, China, Zhu et al. [10] reported on the atomization on a relatively large system scale. Egorov [20] proposed an algebraic
characteristics of a water jet in supersonic air cross flow. The VOF-Level interface area density (AIAD) method that blends the contributions of
set interface tracking and LES turbulence model were combined, with different interface morphologies into the interfacial area density and
7.7 million cells employed to represent a zone measuring 6 mm. drags force, thereby removing any necessity to resolve a specific inter­
Due to the limited computational resources available during prac­ face. The morphologies are divided into three categories: continuous
tical applications, the development of an effective numerical simulation surface, bubbles and droplets; each with its own definitions of interfacial
method for the breakup and atomization of liquid jets has attracted area density and drag coefficient. The AIAD model has been applied by
much attention. Over past decades however, it was the multi-scale-based both [21] and [22] in their simulations of horizontal stratified
hybrid methods that were widely used. Depending on the handling of gas–liquid flows and through these it has been shown to be able to
the dispersed liquid droplets, these multi-scale methods can be classified predict different flow regimes. For the Eulerian framework, the concept
into either Lagrangian or Eulerian approaches. The former uses a of a generalized numerical method of two-phase flow (GENTOP) was
Eulerian scheme to resolve the core liquid jet and accounts for the also raised by [23–25] in order to simulate the behavior of dispersed
dispersed droplets by calculating their respective trajectories. Herrmann droplets or bubbles after they had been detached from a continuous
[11] and Ham et al. [12] used a level set method combined with a LES surface. In this concept, the subsequent coalescences or breakup of the
turbulent model to resolve all the interface dynamics and physical dispersed phase are accounted for by many population balance models
processes that occur at a larger scale (larger than grid resolution). The including ABND [26], MUSIG [27,28] and DQMOM [29,30]. In this way,
small sub-grid scale phenomena were accounted for by a Lagrangian GENTOP can be employed to simulate the evolution of secondary
framework. Tomar et al. [13] and Vallier [14] performed a VOF coupled droplet sizes in liquid jet atomization.
with a Lagrangian particle tracking method (LPT) in order to carry out There are many studies on the multi-scale numerical calculation
the multi-scale simulations of primary atomization and cavitation model of gas–liquid two-phase flow, such as the research carried out in
breakup. The structures used in the continuous interface and larger recent years on AIAD [20–22] and IMUSIG [27,28]. However, these
droplets, or bubbles, were resolved by VOF, while those structures that studies are still in their preliminary stages, the accuracy of their calcu­
were smaller than a prescribed threshold were resolved by LPT. Even lations is not satisfactory, and there remains a lack of knowledge on the
using the above Lagrangian to dispose of the dispersed liquid droplets is relevant experimental verification processes.
computationally cheaper than employing the DNS method to carry out In the present study, which learns from and combines AIAD and
the numerical prediction of primary atomization, though the cost could IMUSIG models, a three-fluid Eulerian approach is proposed and three-
increase exponentially as the number of tracked particles increased. dimensional numerical simulations are performed. This is done in order
Another drawback is the criterion, which determines the transformation to model a liquid jet atomization in which the transition models of both
from the Eulerian to the Lagrangian phase, is commonly a prescribed continuous and dispersed liquid phases are coupled with the droplet
size with respect to the grid size. A comparison between DNS and breakup and coalescence models of the dispersed liquid phase. The
Eulerian-Lagrangian spray atomization (ELSA) made by [15] showed dispersed droplets are regarded as a third Eulerian phase, whose size
that the ELSA method can provide satisfactory results when in the dense distribution is further resolved through a population balance model,
phase zone close to the nozzle. Li and Soteriou [16] adopted a VOF to which thus forms a multi-fluid and multi-scale method. The disinte­
discrete particle model (DPM) method in order to investigate the in­ gration rate of the droplets coming from the continuous liquid was
fluence that density ratio has on the atomization characteristics of liquid realized by a turbulent breakup sub-grid model stemming from flow
jets in cross flow. Some of the liquid lumps belonging to the Eulerian instability. By applying a coarse grid and a Reynolds stress turbulence
phase must be considered when calculating the Sauter mean diameter model (RSM), the three-fluid multi-scale numerical method was able to
(SMD). This is because based on the grid size, some droplets will be be conducted with a much lower computational cost. For validation
unable to be transformed into the Lagrangian phase according to the purposes and a quantitative comparison, the numerical results are pre­
criterion, despite an adaptive mesh method having been used to pre­ sented from different experimental cases found in the literature, with
cisely refine the grid. these results revealing good agreement in terms of droplet size distri­
The general model used to simulate the flow of the liquid–gas is the bution and jet spreading morphology. This method can be considered
Euler Lagrangian discrete droplet method (DDM). This method is the main innovation put forward in this paper. In addition, however, the
numerically unstable [17] and is sensitive to the initial turbulence level comparison between the calculated results and the experimental results
[18]. Moreover, in this model each cell requires a sufficient number of also complements the experimental verification of the two-phase flow
spray parcels if they are to provide a reliable description of the droplet simulation method.
size in the whole computational domain. Thus, due to these issues, a
high computational complexity will be generated. In addition, existing 2. Basic Eulerian equations
multiphase models usually display typical mathematical problems, such
as their non-uniqueness of jumping conditions and the loss of hyperbola. 2.1. Conservative formulations
In order to avoid these difficulties, this paper applies the Euler-Euler-
Euler multi-fluid method to simulate jet atomization. Under the framework of Reynold Averaged Navier-Stokes (RANS),
Meanwhile, because of its short calculation time and its ability to the Eulerian approach deals with the averaged flow fields by assuming
deal with a solid phase as a continuous medium, the Euler-Eulerian that the two phases can interpenetrate each other without having to
multi-fluid model is commonly used for gas–liquid-solid multiphase construct interfaces. The effect of the lost interfaces during the aver­
flows, including dense particle flows [19]. At present, no research that aging process must be accounted for through constitutive relations
studies jet atomization using this method has been found. describing the interactions between the phases. For an isothermal two-
When the droplets are deemed enormous, the Eulerian disposal of the phase flow, the transient form of the mass and momentum conserva­

2
X. Qu et al. Applied Thermal Engineering 195 (2021) 117160

tion equations for continuous gas (cg), continuous liquid (cl) and
dispersed liquid (dl) can be represented as follows: 2.3. Momentum interactions between phases
∂αk ρk
+ ∇(αk ρk Uk ) = Sk (1) Since the dispersed liquid actually departs from the continuous
∂t
liquid, the influence of the droplet momentum on the flow of the
∂αk ρk Uk continuous liquid is neglected in the present three-fluid Eulerian phases
+ ∇(αk ρk Uk ×Uk ) = − ∇(αk pk ) + αk ρk g + ∇αk (τk
∂t modelling. Therefore, two pairs of interactions between two of the three
+ τTk ) + Mk + SMk (2) phases are considered; these being the interactions between gas and
dispersed liquid and the interactions between gas and continuous liquid.
where the subscript k can be cg, cl and dl; τ and τT stand for molecular
viscous stress and turbulence flux respectively; α is the volume fraction 2.3.1. Momentum interactions between gas and dispersed liquid
and αcg + αcl+αdl = 1 since the void fraction of the three fluids sum to The momentum interactions between continuous gas and dispersed
unity; U is the velocity vector; ρ is density; g is gravitational accelera­ liquid are accounted for by the particle model, which treats the
tion; and p is the pressure shared by the two fluids. Finally, Mk is the dispersed liquid as a sphere with varying sizes. Practically, when
interfacial momentum transfer source, i.e. the generalized momentum modeling the liquid particles being injected into the gas space, gas is
exchange vector between phases, and it can be expressed as the sum of taken as the continuous primary phase and liquid is taken as the
several components: dispersed secondary phase, to which a scale or diameter must be
assigned. For the dispersed droplets with an equivalent diameter d, the
Mk = (MD + ML + MTD )k (3)
three momentum exchanges expressed as the volumetric force density
The drag force MD, lift force ML and turbulence dispersion force MTD can be respectively defined as:
are all involved in this study. Sk represents the mass transfer rate, and ⃒ ⃒(
CD )
SMk describes the transfer of momentum between different velocity Mdl,D = − Mcg,D = 0.75 αg ρl ⃒Ug − Ul ⃒ Ug − Ul (9)
d
groups when droplets are forced to switch from one to another velocity
group due to droplet breakup or coalescence processes. ( )
Mdl,L = − Mcg,L = αg ρl CL Ug − Ul × ∇ × Ul (10)

2.2. Turbulence modelling Mdl,TD = − Mcg,TD = − CTD ρl kl ∇αl (11)

In order to save on computational sources, a relatively coarse grid where CD, CL and CTD are drag force coefficient, lift force coefficient and
has been adopted in the present RANS-based method. In the following turbulent dispersion coefficient respectively.
section 3, the sub-grid model used to calculate the disintegration rate Eq. (11) is the classic Lopez de Bertodano model [33], where the
from continuous liquid to dispersed liquid is shown to require a suffi­ value of CTD depends on the specific flow and is set to be 0.5. The Ishii-
ciently accurate representation of turbulence, in terms of its velocity Zuber model was used to obtain CD, since it is suitable for both sparsely
fluctuations and eddy size. RSM abandons the isotropic assumption of and densely dispersed droplet clusters, and it is adapted to a variation of
turbulence, and establishes the model equation of each Reynolds stress. droplet shapes including spheres, ellipses and cap [34]. The Saffman Mei
In addition, this method demonstrates better adaptability for a strong model [35] meanwhile, is a popular model used to calculate the lift force
swirling flow [31], and thus the RSM turbulence model is employed to coefficient CL, and thus it is adopted in the present paper.
close the RANS equations for turbulent terms. The RSM model is the
strictest type of RANS-based model able to reflect the velocity fluctua­ 2.3.2. Momentum interactions between gas and continuous liquid
tions in each direction by solving the Reynolds stresses. Furthermore, The interface tracking methods including VOF and level set were
the extra computational cost required by RSM over the two-equation frequently employed when calculating the mixing flow between two
turbulence models can well be compensated for by the improvement continuous streams. They are able to predict the specifics of the interface
in accuracy [32]. topology, while requiring no further constitutive models for the mo­
The Reynolds transport equation derived from the pulsation equa­ mentum transfers across the interface. However, extremely fine
tion is as follows: computational grids, and thus huge computer resources, are necessary to
∂ ( ’ ’) ∂ ( ) achieve precise resolutions of the interface and the consequent inter­
ρui uj + ρuk u’i u’j = Dij + Pij + εij + Fij (4) phase interactions. In order to compromise between accuracy and
∂t ∂xk
computer consumption, the AIAD model was proposed so as to blend the
The two terms on the left of Eq. (4) represent the Reynolds stress time different interaction mechanisms exerted by different interface topol­
rate of change and the convective term respectively. The five terms on ogies [20].
the right meanwhile are respectively: The basic idea of the AIAD model is to define the blending functions
Turbulent diffusion term: depending on the volume fraction of the continuous gas and liquid
∂
[
( ) ∂ ’ ’
] phases, which then enables switching between dispersed droplets,
Dij = − ρu’i u’j u’k + ρ δkj u’i + δik u’j − μ (ui uj ) (5) dispersed bubbles and the continuous surface. According to these
∂xk ∂xk
blending functions and depending on the local morphology, different
Stress production term: equations can be adopted for the coefficient of momentum interaction
between different morphologies. The blending functions proposed by
∂uj ∂ui
Pij = − ρ(u’i u’k + u’j u’k ) (6) [21] for regions of droplets, bubbles and free surfaces are respectively
∂xk ∂xk
expressed as:
Buoyancy generation term:
fD = [1 + eaD (αcl− αD,limit ) − 1
] (12)
Fij = − 2ρ Ωk (u’j u’m ikm
ε + u’i u’m jkm )
ε (7)

Dissipation term: fB = [1 + eaB (αcg− αB,limit ) − 1

] (13)

∂u’ ∂u’ fFS = 1 − fD − fB (14)

εi,j = − 2μ i j (8)
∂xk ∂xk
where aD and aB are the blending coefficients for droplets and bubbles,

3
X. Qu et al. Applied Thermal Engineering 195 (2021) 117160

respectively and αD,limit and αB,limit are the respective volume fraction the surface. If the turbulent kinetic energy of a surface eddy plus the
limiters. The parameter studies indicate that the result is not very sen­ aerodynamic forces exerted by the flowing gas can overcome the surface
sitive towards these coefficients and limiters [21]. Therefore, aD = aB = energy, a droplet with the size of the eddy can then be disintegrated out
70 and αD,limit = αB,limit = 0.3 are used in the present study, as indicated of the continuous surface. Based on [38], the size of the eddy is the
by [22]. smallest possible droplet that can be formed and so, consideration of
Differently to most interface tracking methods, where only one group smaller droplets is unnecessary. According to this theory, the following
of momentum equation is shared between the two streams (i.e. contin­ criterion was used to judge if a single droplet can be ruptured from the
uous liquid and gas), the AIAD model, which deals with two Eulerian surface of a continuous surface:
fluids, takes care of each fluid and their own momentum fields. Thus, the ( )
momentum interaction (drag force) is induced at the interface. CR = ρcl v2fl + Ca ρcg u2g λ3 − Cs σλ2 > 0 (20)
Regardless of the specific interface geometry, the volumetric drag force
density between the continuous liquid and gas can be expressed as: where Ca and Cs are empirical constants. vfl meanwhile, is the velocity
⃒ ⃒ fluctuations that run vertical to the surface of the continuous liquid. It
|Mcl | = ⃒Mcg ⃒ = 0.5CD Aρ|U|2 (15) can be modeled by the projection of the velocity fluctuation vector in the
direction of the surface between the continuous liquid and gas as follow:
where CD is the dimensionless drag coefficient; A is the interfacial area
density; U is the relative velocity; and ρ is the volume fraction-weight vf l = (u’ ∙n)∙n (21)
averaged density of the two phases.
The magnitude of the shear stress at each side of the interface is the where n is the interface normal vector pointing outward of the fluid, and
module of the product of the viscous stress tensor T and the surface u’ is composed of three anisotropic fluctuation components, which are
normal vector n: obtained from the RSM turbulence model. The term λ is the Taylor
micro-scale, which is the turbulent length scale located in the inertial
τw = |T∙n| (16) sub-range of turbulence energy cascade. It is smaller than the largest
eddies in the flow but larger than the Kolmogorov length scale, which is
where the viscous stress tensor T can be obtained by the Newton inner
in accordance with the assumption by [40] that the critical eddy size is
friction law. It has been assumed by [36] that the shear stresses near the
in the inertial range. The eddy size λ is modeled by turbulent kinetic
interface behave as a wall shear stress on both sides of the interface in
energy k, viscosity ν and dissipation rate ε as follow [41]:
order to reduce the velocity difference of both phases, and thus form the
√̅̅̅̅̅̅̅̅̅̅̅
drag force. Therefore, the drag force in Eq. (15) is equal to: k
λ = 10ν (22)
ε
0.5CD Aρ|U|2 = τw A (17)
As long as the CR computed from Eq. (12) is evaluated to be positive,
Considering the interpenetration of the two phases, in order to get
droplets can be extracted from the continuous liquid phase and released
the volumetric drag force density, the shear stresses should be averaged
into the dispersed liquid phase in each computational cell on the
over the two phases based on the volume fraction:
interfacial regions (where the volume of continuous liquid is smaller
2(αcl τw,cl + αcg τw,cg ) than 1 and greater than 0). However, in order to control the mass con­
CFS = (18)
ρ|U|2 servation in the system, the amount of mass extraction from the
continuous liquid phase must be accurately defined. This value is both
The total drag coefficient for a unit volume is calculated as the sum of temporally and spatially dependent on the solution parameter. In the
the drag coefficients for all morphologies, weighted by the blending present study, we consider a constraint developed by M. Saeedipour
functions: [32] to adjust the sub-grid model, where the production rate of the
CD = fD CD + fB CB + fFS CFS (19) droplets is correlated to the grid spacing and the frequency of fluctua­
tions. The droplet production rate (Ṅ) is modeled as a function of tur­
where in the AIAD model the drag coefficients for bubble and droplet bulence frequency (fr), eddy size (λ) and grid spacing (Δx) as follows:
are both considered to be the same as those of the small sphere solid
particles, which is CD = CB = 0.44. Δx
Ṅ = fr × ( )2 (23)
λ
3. Modeling the transformation between phases
where fr is the frequency of turbulence equal to dividing the fluctuating
velocity vfl by the eddy size λ:
3.1. Disintegration of droplets from continuous liquid
vfl
fr = (24)
In the interface tracking method, where the continuous liquid and λ
dispersed liquid are actually both resolved by fine meshes, no further This sub-grid model will be applied to the interfacial cells with
discrimination of dispersed droplets from continuous surface is neces­ positive CR values and continuous liquid, and then added to the
sary. This is due to the fact that the surface rupturing process to form dispersed liquid. Thus, the additional source terms that appeared in Eq.
isolated droplets is intrinsically linked to the detailed interface evolu­ (1) read:
tion. However, its computational demand is enormous, and with the
added complication of interface topology, it further increases still. Ef­ Sdl = − Scl = ρcl Ṅ
πλ3
(25)
ficiency and accuracy must be compromised on for engineering-oriented 6Δx3
investigations, meaning that a coarser grid is more desirable, as is the The above modelling on the disintegration of droplets from contin­
case in the present study. Since coarse grids are not precise enough to uous liquid was established within the framework of the User Defined
produce an exact interface structure, the rupture of the dispersed droplet Functions (UDF) of Fluent 19.0. Further details regarding the imple­
from the injected liquid is accounted for by the following sub-grid mentation and modification of these source terms in Fluent can be found
model. in [42].
According to [37–39], the breakup or rupture of a liquid jet that is
flowing inside of gas occurs as a consequence of an unbalanced situation
between the energies that are competitively disrupting and stabilizing

4
X. Qu et al. Applied Thermal Engineering 195 (2021) 117160

3.2. Breakup and coalescence of disintegrated droplets ∑

DB = ρg αg fi Bij (32)
j<i
After the rupturing of the droplets at the surface between the
continuous liquid and gas, the adoption of a discrete population equi­ ( )2 ∑ 1
librium model allows for the secondary breakup and coalescence after Dc = ρg αg Cij fi fj (33)
j
mj
collision. As shown in Fig. 1, the droplets were divided into multiple
groups based on the sizes according to the whole span of the size dis­ where Bij and Cij are the breakup rate and coalescence rate between size
tribution. A total of G size groups of droplets is included for the dispersed group i and size group j respectively; m denotes the mass of a droplet in a
liquid phase. Each of the size groups represents a range of droplet sizes, size group; and Xjki is the mass fraction transferred from size group j and
and the droplet breakup and coalescence occur among them. This pro­ k into the droplet when a droplet of size group i is formed due to coa­
cess is illustrated through Fig. 1 By introducing size fraction fi as the lescence. When mj + mk is less than mi-1 or larger than mi+1, Xjki is set to
volume fraction of a particular size group that is fi = αi/αdl, where i = zero and Xjki is set to change bilaterally and linearly from one (when mj
1… G, the continuity equations for each size group in the dispersed + mk = mi) to zero (when mj + mk = mi-1 or mj + mk = mi+1).
liquid phase is: The breakup rate Bij and coalescence rate Cij can be related to ve­
∂ρdl αdl fi locity, turbulence and droplet size. Due to a lack of breakup and coa­
+ ∇(ρdl αdl fi Udl ) = Si (26) lescence mechanisms for droplets, the traditional mechanisms for
∂t
bubbles in the two-phase flow have been used as an elementary sub­
with the additional constraints being: stitute in the present paper. Based on the isotropic turbulence and
probability theory, the Luo and Svendsen model [43] is used to obtain
∑G ∑
G
αdl = αi and fi = 1 (27) the breakup rate, while the coalescence rate is obtained by the Prince
and Blanch model [44]. Future efforts need to focus on the development
i=1
i=1

∑G of more targeted models for droplet swarm evolution following

Sdl = − Scl = i=1
Si (28) disintegration.
In summary, the above models used in the three-fluid Eulerian
Eq. (26) indicates that all the size groups belonging to the same method are illustrated in Fig. 2, along with a schematic of the atomizing
dispersed liquid phase share the same velocity field Udl, which thus liquid jet.
neglects the influence that bubble size change has on the velocity field in
this phase group. 4. Results and discussion
The term Si in Eq. (26) represents the net rate at which mass accu­
mulates in size group i of the dispersed droplet phase. As no phase This section presents the simulation results using the methodology
change is considered in the present study, the value of this term is dictated in this study. Two groups of experimental results from the
determined by the net summation of the droplet breakup and coales­ referenced material were chosen as the validation cases. In the first
cence. Droplet size index i represents the droplet size group, while an group, water was ejected into the atmosphere at different flow rates,
increasing value of the index means an increasing size of droplets. The while the second group ejected ethanol into a pressurized chamber. All
right term of Eq. (26) can be expressed as: the experimental cases were performed using a liquid injection through
Si = (BB + BC − DB − DC )i (29) a circular nozzle into stagnant gas. According to the instability analysis,
the cases covered in the present study are all located in the atomization
where BB is the droplet birth rate due to a breakup of droplets larger than regime.
the i-th size group; BC is the droplet birth rate due to a coalescence of The simulation of the liquid ejection was carried out using the
droplets smaller than the i-th size group; DB is the death rate due to a commercial CFD software package, ANSYS FLUENT 19.0. Though the jet
breakup of the i-th size group into smaller droplets; and DC is the death itself demonstrates quasi-axisymmetric features around its nozzle axis, a
rate due to a coalescence of i-th size group with other size groups. To be three-dimensional simulation domain must still be adopted in order to
specific, the four items on the right of Eq. (29) respectively possess the capture the anisotropic turbulence, which determines the phase
following forms: discrimination in the present methodology. The calculation domain’s
∑ geometry is shown in Fig. 3, alongside the boundary conditions. No-slip
BB = ρ g α g Bij fj (30) wall conditions are applied on the four sides and the top of the rectan­
gular solid domain. Constant pressure is applied to the outlet, while
j>i

( )2 ∑∑ mj + mk constant velocities, depending on the different cases, are applied to the

Bc = 0.5 ρg αg Cjk fj fk × Xjki (31) liquid inlet located on the top wall of the calculation domain.
mj mk
A computational region was chosen which is large enough to prevent
j≤i k≤i

fluid breakup and stop the velocity field of the surrounding gas from
being influenced by the boundary [45]. For the other cases, the grid size
was determined in the same way. The fluid inlet is located in the center
of the mesh and the wall boundary condition is set near the nozzle. The
pressure outlet boundary condition is on the other side. The SIMPLE
algorithm is used to solve the coupled pressure–velocity problem for
both the momentum and continuum equations. The advanced k-ζ-f
model [46] is used to solve the Reynolds-averaged Navier-Stokes
(RANS) equations. The minimum time step is 1e-8, the maximum time
step is 5e-6, and the CFL number is 4. Ten Eulerian phases are defined:
one gas phase, one liquid phase and nine droplet phases. The second
order upwind scheme is used for droplet phases while the second order
implicit scheme is used for the transient term. The convergence criterion
is that the equation’s residual is less than 10-3.
Both the Sauter mean diameter (SMD) D32 and number-mean
Fig. 1. Extension of the discrete population balance model.

5
X. Qu et al. Applied Thermal Engineering 195 (2021) 117160

Fig. 2. Schematic of an atomizing jet and sub-models used in three-fluid Eulerian method.

and atomization characteristics. Two liquid mass flow rates, which are
denoted as case 1 and case 2, were simulated with the corresponding
velocities of 45 m/s and 60 m/s respectively and the respective Reynolds
numbers 45,000 and 60125.
In the experiment, phase-Doppler anemometry (PDA) was adopted in
order to measure the local droplet’s size and velocity. As the beams
emitted from the PDA system are unable to penetrate areas with a dense
interface, only the spray features that were far downstream from the
nozzle (200 mm, 300 mm and 400 mm) were provided in the test.
According to the experiment, the droplet sizes are distributed in the
range of 0 um to 160 um with an SMD less than 100 um. The water
Fig. 3. Geometry of calculation domain.
velocity distribution and droplet size for D10 at the axial position of 200
mm were selected to test the grid independence and the results are
diameter D10 are important indexes for the two-phase flow and are used shown in Fig. 4. As there were a high number of droplets during the
as comparison quantities between the simulation and experiment. SMD atomization phase, the hexahedral mesh was used. It can be seen that the
is a diameter of a virtual droplet whose volume-to-surface ratio is the simulated values with 3,123,658 and 2,056,937 grid numbers are
same as the ratio of total droplet volume to total droplet surface. The basically coincident, indicating therefore that when the number of grids
number-mean diameter is an averaged diameter weighted by the num­ reaches a certain amount, the number of grids increases but it has little
ber density of droplets. The definitions for D32 and D10 are expressed by effect on the result. Therefore, 2,056,937 grids are selected for the
Eq. (34) and (35), respectively: calculation and a structured mesh with the minimum grid space of one-
∑ 3
ni d eighth of the nozzle diameter, which is slightly larger than a typical
D32 = ∑ i2 (34) droplet, is generated by ANSYS ICEM. 53 × 53 × 500 divisions are
ni di
generated in the three directions of the calculation domain (100 × 100
∑
ni di × 500 mm) as shown in Fig. 3, which results in a total number of
D10 = ∑ (35) 2,056,937 mesh elements.
ni
In order to test group independence, the number of groups was
where ni is the number density of droplets with diameter di, which can explored. The water velocity distribution and droplet size for D10 at the
be related to the size fraction fi and dispersed liquid volume fraction αdl axial position of 200 mm in 6, 9 and 12 groups were therefore selected.
as: The results are shown in Fig. 5. In the atomization phase, it can be seen
that the simulated values when there are 9 or 12 groups are basically
6fi αdl
ni = (36) coincident, indicating therefore that when the number of groups reaches
πdi3
a certain amount, it has little effect on the result. Therefore, 9 groups are
selected for calculation.
4.1. Results for the water ejection In the simulation, nine size groups, from 10 um to 160 um with an
increasing ratio of 1.41, are adopted to represent the dispersed droplets
The water–air two-phase ejection flow mentioned in the two exper­ which have ruptured from the continuous liquid phase. The breakup and
imental cases by [47] was reproduced using the three-fluid Eulerian coalescence models are thus activated for the ruptured droplets, mean­
numerical method. In the investigation, water and air were taken at ing that the size and number of the droplets in the dispersed phase varies
standard room temperature with the following properties: ρl = 1000 kg/ along the ejection axis.
m3, ρg = 1.2 kg/m3, μl = 1 mPa⋅S, μg = 0.0017 mPa⋅S and σ = 0.072 N/ In the models shown in section 3, the transformation of continuous
m. liquid into dispersed liquid occurs where the turbulent energy carried by
In order to find the proper Reynolds number to ensure that the fluid parcels exceeds the surface energy. This process is illustrated in
breakup regime is atomized, a nozzle with a length-to-diameter ratio of Fig. 6 showing the sequence of turbulence kinetic energy, the velocity
10 and a diameter 1 mm was used in the experiment, while relying on fluctuations and the disintegration rate. As expected, the result of the
the v. Ohnesorge’s nomogram [48]. This apports some difficulties when shear stresses produced between the high-speed jet and the stagnant gas
determining the turbulence velocity boundary condition of the liquid means that the turbulence is most intensive near the contact point of the
inlet. Though the flow may not be fully developed in the short tube liquid and gas. Both the turbulence kinetic energy and the velocity
before the nozzle exit, a turbulent intensity of 0.05 was still applied. It is fluctuations demonstrate their maximum values near the liquid ejection
worth noting that this may result in deviations in the downstream spray region, while the values gradually decrease along the flow direction. The

6
X. Qu et al. Applied Thermal Engineering 195 (2021) 117160

100 80

80 Case1-1682312 60 Case1-1682312
Case1-2056937 Case1-2056937
Case1-3123658 Case1-3123658

60 40
D10 (um)

U (m/s)
40 20

20 0

0 -20
0 10 20 30 40 50 0 10 20 30 40 50
r (mm) r (mm)

Fig. 4. The distributions of droplet size and velocity at the axial position of 200 mm for different meshes.

100 50

Case1-6 bins 40
80 Case1-6 bins
Case1-9 bins Case1-9 bins
Case1-12 bins 30 Case1-12 bins
60
20
D10 (um)

U (m/s)

40 10

0
20
-10

0 -20
0 10 20 30 40 50 0 10 20 30 40 50
r (mm) r (mm)

Fig. 5. The distributions of droplet size and velocity at the axial position of 200 mm in 6, 9, and 12 groups.

disintegration rate that turns the continuous liquid into dispersed liquid shown in Fig. 7, the simulation overpredicts, to a very large extent, the
greatly depends on the turbulence, as indicated by Eq. (30), and thus this increment in the droplet size from the 200 mm cross-section to the 400
shows the same trend as the velocity fluctuations. Based on the simu­ mm cross-section. The simulation overpredicts the D10 but then under­
lation, the disintegration rate is found to increase, due to the increase in predicts the D32. It is also evident that the simulation predicts the D10 in
the ejection velocity and the turbulence intensity, from 175 kg/m3/s for the x = 200 mm much more accurately than it does in the x = 400 mm.
case 1 to 310 kg/m3/s for case 2. In contrast, the D32 in x = 400 mm are predicted in a more accurate
The droplet size radial distributions, which are obtained from the manner than those in x = 200 mm. In other words, the simulation can
simulation, are compared with the PDA measurements in Fig. 7. From give acceptably accurate D10 in the cross-sections near the liquid ejec­
the simulation, both D10 and D32 are found to first decrease from their tion and acceptably accurate D32 in the cross-sections that are far from
maximum value when in the ejection axis and then increase after the liquid ejection. This creates a dilemma when adjusting the droplets’
reaching their minimum. This decrease and increase variation tendency collision coefficients in the coalescence model to try and to improve the
seems to have been missed by the experimental measurements, but this accuracy. For instance, if the coalescence rate increases when using
is mainly because only a small radial range was covered by the PDA test. larger collision parameters, the droplets size at x = 200 mm will increase
It seems that the measured variations of D10 happen to be located in the and thus the deviation of the predicted D32 will be reduced, however,
increasing segment of the simulation, while the measured D32 variations then the deviation of the predicted D10 will increase. Vice versa, if the
conform mostly to the decreasing segment of the simulation. Although coalescence rate decreases when using smaller collision parameters, the
only a small radial range was covered by the experiment and isolated droplets size at x = 400 mm will decrease and thus the deviation of the
data points are obtained by PDA, these discrete data points exhibit predicted D10 will be reduced, however, the deviation of the predicted
reasonable compliance with the simulation’s data line. As the jet evolves D32 will increase. In order to improve the accuracy of the present three-
downstream, larger-sized droplets are obtained from both the simula­ fluid Eulerian method then, further numerical efforts must be invested in
tion and the experiment, which indicates that the coalescence between exploring the appropriate model parameters, which includes the burden
droplets overwhelms the possible breakup and hence the droplets’ of carrying out numerous trial calculations.
diameter increases streamwise, as shown by the comparison between the Fig. 8 compares the droplet velocities in the × direction at different
D10 and D32 at the axial positions of 200 mm and 400 mm. axial locations. Both the experiment and simulation exhibit a decreasing
Despite the possible measurement uncertainty by PDA, a noticeable trend from the ejection axis towards its periphery. The simulation’s
discrepancy still exists between the simulation and experiment. As predicted velocity agrees reasonably well with the experimental data at

7
X. Qu et al. Applied Thermal Engineering 195 (2021) 117160

Fig. 6. Distributions of turbulence and disintegration rate (case 1).

the radial location close to the axis (less than 5 mm) while the 4.2. Results for the ethanol ejection
discrepancy starts to enlarge as the measurement points move outwards.
However, for case 1, the measurement gives higher droplet velocities Ethonal and N2 at 1 MPa were taken in the investigation with the
when going downstream while the simulation gives higher droplet ve­ following properties: ρl = 790 kg/m3, ρg = 10.1 kg/m3 , μl = 0 . 0011 Pa
locities when going upstream. This disagreement, as well as the s, μg = 17.89 × 10-6Pa⋅S and σ = 0.021 Nm. The two cases shown in
enlarging discrepancy toward the jet periphery, may be attributed to the section 4.1 are for liquid ejection into gas under an atmospheric pres­
combined effect of the narrow variation of the droplet velocities be­ sure, with the density ratio of liquid to gas being more than 800. In
tween these two axial locations and the comparatively large PDA mea­ scenarios such as the ejection of fuel into a combustion chamber or the
surement errors. cooling of a nuclear Reactor Pressure Vessel (RPV) through spraying, the
A bonus offered by the Eulerian description of the dispersed droplet gas is pressurized to a very large value which means that the gas density
phase is that the distribution of the volume fraction of the different increases considerably. Therefore, in order to gain basic knowledge on
droplet size groups (size fraction) can be conveniently obtained. The the applicability of the method for low density ratio ejection and at­
distributions of five (10 um, 20 um, 40um, 80 um and 160 um) out of the omization, one of the experimental cases (denoted as case 3 in the
total nine size fractions are demonstrated in Fig. 9. As expected, the present study) from the work carried out by reference [32] is tested.
smallest droplet size prescribed in the model is concentrated near the The shape of the calculation domain and the boundary conditions are
nozzle ejection while the larger droplet size tends to aggregate down­ the same as those shown in Fig. 3, but with the dimensions of 12 × 12 ×
stream, thereby indicating that the coalescence between droplets over­ 90 mm. The smallest grid size is also designed as slightly larger than a
whelms the possible breakup downstream, which is consistent with typical droplet size, and is also distributed between 10 um and 160 um.
Fig. 7. It is worth noting that the value shown in Fig. 9 is the fraction of For the ejection of ethanol into a high-pressure chamber, an exper­
each droplet size in the total dispersed droplet phase, and all nine imental correlation was offered by [38] to predict the evolution of SMD
fractions sum to a unit. along the nozzle axis, which can be expressed as follows:
The global Sauter mean diameter at the three downstream vertical ( ) ( )34 ( 5 )
planes are shown in Fig. 10, and are compared with the experimental σ ρg We6
(37)
1
D32 = 0.25 x3
and numerical results by [47]. The D32 increases with the distance from ρg U02 1
ρl Re2
the liquid ejection, indicating therefore that the droplet coalescences
overwhelm the droplet breakup after disintegrating from the continuous where U0 is the ethanol ejection velocity; x is the axial distance from the
liquid. The predicted global SMD for case 2 is observed as being clearly nozzle exit; and We and Re are respectively the liquid Webber number
smaller than that of case 1 because even when the eddy size is small, the and Reynolds number defined with a characteristic length of one-fourth
enhanced velocity turbulence stimulates more surface rupture, as indi­ of the nozzle diameter. In Fig. 11, a Eulerian-Lagrangian method is used
cated by Eq. (20). This is also in accordance with the experimental ob­ to compare the SMD from the present simulation with those from Eq.
servations. Conclusively, both the present three-fluid Eulerian method (37), as well as with the numerical results. It can be concluded that the
and the Eulerian-Lagrangian method given by [47] can offer numerical present method can predict the SMD evolvement reasonably well.
results that have satisfactory agreement. The probability density function (PDF) of the droplet diameter at a
distance of x/d = 20 is plotted in Fig. 12, together with that from the
simulation carried out by [32]. Since there is no need to designate the

8
X. Qu et al. Applied Thermal Engineering 195 (2021) 117160

(a) D10 for case 1 (b) D10 for case 2

(c) D32 for case 1 (d) D32 for case 2

Fig. 7. Comparisons of radial distributions of the number-mean droplet size D10 and Sauter-mean droplet size D32 at different axial locations.

(a) case 1 (b) case 2

Fig. 8. Comparisons of the radial distributions of the x-component droplet velocity.

droplet size in the Lagrangian method and the droplet size disintegrated therefore conclude that with regards to droplet size distribution, when
from the surface continuously varies, the PDF from this method can be the gas is pressurized, the present three-fluid Eulerian method can still
drawn in quite a narrow diameter interval. In contrast, a total of nine give satisfactory liquid jet atomization results.
droplet sizes are used in the present three-fluid Eulerian method. This
means that the interval in the PDF for the current simulation depends on 5. Conclusion
the diameter prescription and is unequal, as illustrated in Fig. 12.
Nevertheless, the PDF in the present study shows its peak value at Taking continuous gas, continuous liquid and dispersed droplet as
around 55 um, which is in agreement with the PDF obtained by [32] the three independent phases, a three-fluid Eulerian method is proposed
which shows a peak value at around 50 um. From the above, we can for the simulation of liquid jet atomization. Its capabilities are primarily

9
X. Qu et al. Applied Thermal Engineering 195 (2021) 117160

Fig. 9. Size fraction distributions of the dispersed liquid phase (case 2).

Fig. 10. Comparisons of global Sauter mean diameter from simulation and Fig. 11. Comparison of the SMD values with experiment correlations (case 3).
measurement.
adopted due to the increasing computing burden posed by the
explored and tested against several experimental cases, concerning the Lagrangian method as the number of droplets increases. In addition, it is
aspects of flow spread and size distribution of the dispersed droplet adopted because it can conveniently adapt droplet collision and breakup
phase. The Eulerian representation of the dispersed droplets is mainly mechanisms to account for the evolution of the secondary droplet size.

10
X. Qu et al. Applied Thermal Engineering 195 (2021) 117160

characteristic and engine performance in gasoline direct injection engines. Applied

Energy,2018,220.
[2] C. Hirt, B. Nichols, Volume of fluid (VOF) method for the dynamics of free
boundaries, J. Comput. Phys. 39 (1981) 201–225.
[3] S. Osher, J.A. Sethian, Fronts propagating with curvature-dependent speed:
algorithms based on Hamilton-Jacobi formulations, J. Comput. Phys. 79 (1988)
12–49.
[4] M. Sussman, E.G. Puckett, A coupled level set and volume-of-fluid method for
computing 3Dand axisymmetric incompressible two-phase flows, J. Comput. Phys.
162 (2000) 301–337.
[5] M. Gorokhovski, M. Herrmann, Modeling primary atomization, Annu. Rev. Fluid
Mech. 40 (2008) 343–366.
[6] J. Shinjo, A. Umemura, Detailed simulation of primary atomization mechanisms in
Diesel jet sprays (isolated identification of liquid jet tip effects), Proc. Combust.
Inst. 33 (2) (2011) 2089–2097.
[7] K. Hanthanan Arachchilage, M. Haghshenas, S. Park, L. Zhou, Y. Sohn,
B. McWilliams, K. Cho, R. Kumar, Numerical simulation of high-pressure gas
atomization of two-phase flow: Effect of gas pressure on droplet size distribution,
Adv. Powder Technol. 30 (11) (2019) 2726–2732.
Fig. 12. Probability density function (PDF) for droplet diameter at x/d = 20
[8] F.J. Salvador, J.V. Romero, M.D. Roselló, D. Jaramillo, Numerical simulation of
(case 3). primary atomization in diesel spray at low injection pressure, J. Comput. Appl.
Math. 291 (2016) 94–102.
[9] C. Shao, K. Luo, Y. Yang, J. Fan, Detailed numerical simulation of swirling primary
The droplet disintegration rate is designated to relate to the liquid
atomization using a mass conservative level set method, Int. J. Multiph. Flow 89
turbulence. The interactions between gas and continuous liquid are (2017) 57–68.
modeled in the framework of AIAD, while the interactions between gas [10] Y.H. Zhu, F. Xiao, Q.L. Li, R. Mo, C. Li, S. Lin, LES of primary breakup of pulsed
liquid jet in supersonic crossflow, Acta Astronaut. 154 (2019) 119–132.
and dispersed droplets are modeled by a population balance model. The
[11] M. Herrmann, Modeling primary breakup: a three-dimensional Eulerian level set/
following aspects support the capability of the present method: vortex sheet method for two-phase interface dynamics, Annual Research Briefs.
Stanford University, Center for Turbulence Research, pp. (2003) 185-196.
(1) The space distributions of the number-mean droplet size, Sauter- [12] F. Ham, Y.N. Yong, S. Apte, M. Herrmann, A hybrid Eulerian-Lagrangian method
for LES of atomizing spray. Computational Methods in Multiphase Flow II,
mean diameter and droplet velocity can be predicted in reason­ Advances in Fluid Mechanics, WIT Press (2003) 313–322.
able agreement with the experimental measurements. In addi­ [13] G. Tomar, D. Fuster, S. Zaleski, S. Popinet, Multiscale simulations of primary
tion, the discrepancy is considered to be reducible by carrying out atomization, Comput. Fluids 39 (2010) 1864–1874.
[14] A. Vallier, Eulerian and Lagrangian Cavitation Related Simulations Using
improvements on the sub-models, including droplets coalescence OpenFOAM Licenciate of Engineering Thesis, Lund University, 2010.
and breakup. [15] R. Lebas, T. Menard, P.A. Beau, A. Berlemont, F.X. Demoulin, Numerical
(2) As a result of the enhanced turbulence intensity, both the local simulation of primary break-up and atomization: DNS and modelling study, Int. J.
Multiph. Flow 35 (3) (2009) 247–260.
and global droplet size decrease with the increase of the liquid [16] X. Li, M.C. Soteriou, Detailed numerical simulation of liquid jet atomization in
ejection speed. crossflow of increasing density, Int. J. Multiph. Flow 104 (2018) 214–232.
(3) When there are 9 size groups, the larger the droplet size, the [17] Abraham J. What is adequate resolution in the numerical computations of transient
jets?. SAE Tech Pap; 1997. doi: 104271/970051.
higher the value its size fraction tends to reach downstream,
[18] Venkatraman Iyer, John Abraham, Penetration and dispersion of transient gas jets
which indicates that after disintegrating from the continuous and sprays, Combust. Sci. Technol. 130 (1-6) (1997) 315–334.
liquid, the droplets’ coalescence overwhelms the breakup. [19] Numerical and experimental investigation of the spray quenching process with an
Euler-Eulerian multi-fluid model.
(4) When the atmosphere is at a higher pressure or the liquid to gas
[20] Y. Egorov, Validation of CFD codes with PTS-relevant test cases, 5th Euratom
density ratio is low, the method can also produce a satisfactory Framework Programme ECORA project, 91-116. EVOL-ECORA-D07, 2004.
size distribution. [21] T. Höhne, J.-P. Mehlhoop, Validation of closure models for interfacial drag and
turbulence in numerical simulations of horizontal stratified gas–liquid flows, Int. J.
Multiph. Flow 62 (2014) 1–16.
This newly proposed method is a complex fusion of three Eulerian [22] T. Höhne, C. Vallée, Experiments and Numerical Simulations of Horizontal Two-
phases, which includes their interphase interactions, interphase trans­ Phase Flow Regimes Using an Interfacial Area Density Model, Journal of
formation and droplet collision. Therefore, the numerical results pre­ Computational Multiphase Flows 2 (3) (2010) 131–143.
[23] S. Hänsch, D. Lucas, T. Höhne, E. Krepper, Application of a new concept for multi-
sented here only serve as a primary exploration. It is necessary to adjust scale interfacial structures to the dam-break case with an obstacle, Nucl. Eng. Des.
the parameters of the droplet collision model or to develop a new droplet 279 (2014) 171–181.
collision model in order to take care of the secondary atomization of the [24] T. Höhne, E. Krepper, G. Montoya, D. Lucas, CFD-simulation of boiling in a heated
pipe including flow pattern transitions using the GENTOP concept, Nucl. Eng. Des.
droplets and hence improve the droplet size distribution downstream. 322 (2017) 165–176.
To realize this, more experimental measures must be carried out for [25] S. Hänsch, D. Lucas, E. Krepper, T. Höhne, A multi-field two-fluid concept for
validation purposes. transitions between different scales of interfacial structures, Int. J. Multiph. Flow
47 (2012) 171–182.
[26] S.C.P. Cheung, G.H. Yeoh, J.Y. Tu, On the numerical study of isothermal vertical
Declaration of Competing Interest bubbly flow using two population balance approaches, Chem. Eng. Sci. 62 (17)
(2007) 4659–4674.
[27] S. Wang, J. Wen, Y. Li, H. Yang, Y. Li, J. Tu, Numerical Prediction for Subcooled
The authors declare that they have no known competing financial Boiling Flow of Liquid Nitrogen in a Vertical Tube with MUSIG Model, Chin. J.
interests or personal relationships that could have appeared to influence Chem. Eng. 21 (11) (2013) 1195–1205.
[28] X. Li, W. Wei, R. Wang, Y. Shi, Numerical and experimental investigation of heat
the work reported in this paper. transfer on heating surface during subcooled boiling flow of liquid nitrogen, Int. J.
Heat Mass Transf. 52 (5-6) (2009) 1510–1516.
Acknowledgement [29] L. Deju, S.C.P. Cheung, G.H. Yeoh, J.Y. Tu, Capturing coalescence and break-up
processes in vertical gas–liquid flows: Assessment of population balance methods,
Appl. Math. Model. 37 (18–19) (2013) 8557–8577.
The authors would like to extend their sincere gratitude to National [30] S.C.P. Cheung, L. Deju, G.H. Yeoh, J.Y. Tu, Modeling of bubble size distribution in
Natural Science Foundation of China (No. 51806128 and 51879154) isothermal gas–liquid flows: Numerical assessment of population balance
approaches, Nucl. Eng. Des. 265 (2013) 120–136.
and China Postdoctoral Science Foundation (No. 2020M682211).
[31] A. Morrall, S. Quayle, M.S. Campobasso, Turbulence modelling for RANS CFD
analyses of multi-nozzle annular jet pump swirling flows, Int. J. Heat Fluid Flow 85
References (2020) 108652, https://doi.org/10.1016/j.ijheatfluidflow.2020.108652.
[32] M. Saeedipour, S. Pirker, S. Bozorgi, S. Schneiderbauer, An Eulerian-Lagrangian
hybrid model for the coarse-grid simulation of turbulent liquid jet breakup, Int. J.
[1] Tawfik Badawy,Mohammadreza Anbari Attar,Peter Hutchins,Hongming Xu,Jens
Multiph. Flow 82 (2016) 17–26.
Krueger Venus,Roger Cracknell. Investigation of injector coking effects on spray

11
X. Qu et al. Applied Thermal Engineering 195 (2021) 117160

[33] M. Lopez de Bertodano, Turbulent Bubbly Flow in a Triangular Duct, Rensselaer [41] S. Pope, Turbulent Flows, Cambridge University Press, 2000.
Polytechnic Institute, Troy New York, 1991. [42] FLUENT, User Manual. Ansys Inc, in, 2019.
[34] M. Ishii, N. Zuber, Drag Coefficient and Relative Velocity in Bubbly Droplet or [43] H. Luo, H.F. Svendsen, Theoretical Model for Drop and Bubble Breakup in
Particulate Flows, AIChE Journal 25 (1979) 843–855. Turbulent Dispersions, AIChE J. 42 (1996) 1225–1233.
[35] P.G. Saffman, The lift on a small sphere in a slow shear flow, J. Fluid Mech. 31 [44] M.J. Prince, H.W. Blanch, Bubble Coalescence and Break-Up in Air-Sparged Bubble
(1968) 624. Columns, AIChE J. 36 (1990) 1485–1499.
[36] T. Höhne, et al., Numerical simulations of counter-current two-phase flow [45] Ivan Pađen, Zvonimir Petranović, Wilfried Edelbauer, Milan Vujanović, Petranović
experiments in a PWR hot leg model using an interfacial area density model, Zvonimir, Edelbauer Wilfried, Vujanović Milan. Numerical Modeling of Spray
International Journal of Heat and Fluid Flow 32 (5) (2011) 1047–1056. Secondary Atomization with the Euler-Eulerian Multi-Fluid Approach, Comput.
[37] L. Tseng, G.A. Ruff, G.M. Faeth, Effects of gas density on the structure of liquid jets Fluids 222 (2021) 104919, https://doi.org/10.1016/j.compfluid.2021.104919.
in still gases, AIAA Journal 30 (1992) 1537–1544. [46] K. Hanjalić, M. Popovac, M. Hadžiabdić, A robust near-wall elliptic-relaxation
[38] G. Faeth, L. Hsiang, P. Wu, Structure and breakup properties of sprays, Int. J. eddy-viscosity turbulence model for CFD, Int. J. Heat Fluid Flow 25 (6) (2004)
Multiph. Flow 21 (1995) 99–127. 1047–1051.
[39] K. Sallam, Z. Dai, G. Faeth, Drop formation at the surface of plane turbulent liquid [47] M. Saeedipour, S. Schneiderbauer, G. Plohl, G. Brenn, S. Pirker, Multiscale
jets in still gases, Int. J. Multiph. Flow 25 (1999) 1161–1180. simulations and experiments on water jet atomization, Int. J. Multiph. Flow 95
[40] P. Wu, G.M. faeth, Aerodynamic effects on primary breakup of turbulent liquid, in: (2017) 71–83.
Proceedings of the 31st Aerospace Sciences Meeting & Exhibiti (AIAA 93-0903), [48] Wolfgang V. Ohnesorge, Die Bildung von Tropfen an Düsen und die Auflösung
Reno, NV, 1993. flüssiger Strahlen, John Wiley & Sons Ltd 16 (6) (1936) 355–358.

12